Latiai — agentic threat model

Integrates multiple third-party foundation models (Sora, Veo, Flux, etc.). Highly vulnerable to adversarial prompt injections designed to bypass safety filters, leading to the generation of deepfakes, copyrighted material, or policy-violating content.

Not certain from the listing — likely stores user-submitted prompts and generated visual assets. Risks include unauthorized access to user galleries, data exfiltration of proprietary prompts, and lack of clear data retention policies.

Not certain from the listing — orchestration is limited to routing user prompts to the selected model API. Risks include insecure API integration, prompt leakage, and lack of input validation before forwarding to upstream model endpoints.

Not certain from the listing — hosted as a web application. The primary threat is the exposure or theft of high-value API keys used to authenticate with premium upstream model providers (e.g., OpenAI, Google).

Not certain from the listing — no explicit mention of output monitoring or input guardrails. Gaps in observability could allow users to systematically abuse the platform for generating harmful media without detection.

Not certain from the listing — relies on standard web authentication and subscription billing. Key threats include credential stuffing, credit/token theft, and potential non-compliance with emerging AI copyright and safety regulations.

Not certain from the listing — does not feature multi-agent collaboration or marketplace integrations. Threats are limited to dependency on the availability and security of upstream model provider ecosystems.